You are on page 1of 4

Assignment: Data Mining Research Article Group: M. Salman Saeed & M.

Amir Irshad

Suitable Parameter Selection For Regression Analysis To


Predict Natural Gas Transmission Line Pack
Abstract: -

Natural Gas Transmission pipeline networks are back bone of energy supply chain as these
are used to collect natural gas from the natural gas procession plants and fields and
transport it by the help of pipeline network to city gates and consumers. These pipeline
networks has different installations to transport gas including but not limited to regulators,
compressors, valves, filters and measurement skids / gadgets. Optimum level of gas
inventory referred to as ‘Line Pack’ is required in these pipeline networks for smooth
transmittal of natural gas. Most of the installations on these networks have some pressure,
temperature, flow and differential pressure constraints for effective operation and safety.
These constraints define an overall line pack range under which all these constraints are
being tackled effectively. We call it minimum and maximum line pack conditions and natural
gas transmission systems are operated in between these limits for efficiency, safety and
economical operations.

In this case we predicted line pack against total gas input and major consumption sector i.e.
power with linear regression . We predicted line pack against each parameter separately and
then jointly and studied MAE(Mean Absolute Error), MSE(Mean Squared Error) and RMSE
(Root Mean Squared Error). We found out that these errors are lesser when line pack was
predicted against total gas input and gas consumption by power instead of when predicted
separately. However, the errors were large enough to demand a separate study to find better
regression algorithms that may reduce these errors.

Introduction: segments. This pressure drop is then


replenished by compression. Natural gas
Gas transmittal depends upon the
Compressor stations consisted of number
pressure difference created either by
of compressor units are strategically
consumption or by gas compression
placed on transmission network to
which pulls the gas increases its pressure
accommodate for the pressure loss and
which creates the required pressure
further manipulation of gas flow. These
gradient to move forward. Due to internal
compressor stations are thus designed for
boundary wall friction between gas
specified pressure ranges and these are
particles and pipe inner walls, pressure
defined by minimum pressure, maximum
drop occurs across length of the pipeline

Page 1 of 4
Assignment: Data Mining Research Article Group: M. Salman Saeed & M. Amir Irshad

pressure and maximum allowable demand for power will reduce to 400
operating pressure (MAOP). Compressor MMSCFD on a rainy day which is
stations are operated economically and unpredictable 3 month back when
safe between these limits and therefore monthly demand for power was finalized.
these limits are strictly adhered to. Therefore, we need a prediction model to
predict line pack value against changing
When imbalance in gas input and output
power consumption and unscheduled gas
occurs the gas inventory in the
input variations so as to avoid extreme
transmission system is disturbed. This
packing or depletion of line pack.
disturbance is reflected in line pack
depletion or addition. This addition and History and Literature:
depletion in line pack is further reflected in
Andrey A. Belevitin1 ,1 NIIgazeconomika
increase and decrease in system LTD, 20-8, Staraya Basmannaya str.,
pressures at upstream and downstream Moscow, Russia, 105066, in his paper “An
of compressor stations. We do not want Algorithm for Calculating Gas
Transmission System Optimal Stationary
the pressure conditions violating the safe
Mode” developed a model to simulate
operation regime for which we define safe optimum gas input, output and line pack
operating line pack. We understand that conditions for minimum compression. He
natural gas being a compressible fluid, its also used Temperature for additional
criterion for his studies on Russian
inventory / line pack calculation involves
Transmission gas network. However, no
parameters like Pressure, Temperature prediction model was presented for
and Compressibility. Here we are only boundary conditions of line pack as the
considering pressures. Therefore, instead scope was towards optimization.
Zlatan SI.Anica 1, Stjepan SU.I.1, & Boris
of keeping a list of safe pressure ranges at
Mila. (Member, IEEE),Faculty of Electrical
numerous locations, we maintain a range Engineering and Computing, University of
of line pack under which adequate Zagreb, 10000 Zagreb, Croatia discussed
pressures are available for compression about integration of AI based decision
models in SCADA operation to suggest
facilities.
suitable decision in “ Architecture of an
As the variation in demand and supply Artificial Intelligence Model Manager for
Event-Driven Component-Based SCADA
affects the line pack, a prediction model is
Systems”.
necessary to ascertain probable pack
conditions in future with respect to Not much of sort has been tried in
changing power demand owing to Pakistan either in optimizing or integrating
prediction models into SCADA systems of
changing weather forecasts. For example,
gas utility companies. Our study has
a routine demand of 800 MMSCFD gas

Page 2 of 4
Assignment: Data Mining Research Article Group: M. Salman Saeed & M. Amir Irshad

opened the door for future work to be done


for gas transmission networks in Pakistan.

Methodologies:

Linear Regression model is used to predict


line pack against gas input and gas
consumption by power sector. It is
important to note that safe line pack range
is considered to be from 4300 MMSCF to
4900 MMSCF and usually line pack does
not drop below 3900 MMSCF and elevate
above 5200 MMSCF. The differential
between safe limits is 700 MMSCF which
is overshadowed by the overall operating
Performance Evaluation:
range i.e. 4000 MMSCF ~ 5000 MMSCF.
Therefore, to get better results a figure of MAE(Mean Absolute Error), MSE(Mean
3400 is being subtracted from line pack Squared Error) and RMSE (Root Mean
values and similarly a figure of 500 is being Squared Error) are calculated for all three
subtracted from the gas input. cases and values were compared to
ascertain better combination for
Dataset is used 70% for training and 30%
predictions. Graphs were also plotted
for testing. Linear regression is being used
against dates and among parameters to
in python to predict line pack values first
study the behavior and relation among
for input and power separately and then
these.
jointly.
Conclusions:

Values for MAE, MSE and RMSE were high


which suggests either the adoption of
different regression algorithm or further
pre and post data processing.

However, model performed better when


both input and gas consumption by power
were taken jointly to predict line pack.

Page 3 of 4
Assignment: Data Mining Research Article Group: M. Salman Saeed & M. Amir Irshad

For Input only : For Input & Power:

Coefficients: [-1.77457899] ; MAE: Coefficients: [-2.16727709 -0.46167252] ;


280.36986789019306 ; MSE: MAE: 277.53145831550125 ;
122956.00043549144 ; MSE: MSE:115146.90941908969 ; RMSE:
350.6508240907063 339.3330361445665

For Power only: Future Work:

Coefficients: [-0.30153321] ; MAE: This is just the beginning, many


316.19341874584427 ; MSE: improvements can be made by including
138957.78005129218 ; RMSE: Temperature, Pressure and different
372.770411984766 locations for predictions. Polynomial
Regression algorithm should also be tried
and are expected to give better results.

REFERENCES:

1- Andrey A. Belevitin1 ,1 NIIgazeconomika LTD, 20-8, Staraya Basmannaya str.,


Moscow, Russia, 105066, “An Algorithm for Calculating Gas Transmission System
Optimal Stationary Mode”.
2- Zlatan SI.Anica 1, Stjepan SU.I.1, & Boris Mila. (Member, IEEE),Faculty of Electrical
Engineering and Computing, University of Zagreb, 10000 Zagreb, Croatia,
“Architecture of an Artificial Intelligence Model Manager for Event-Driven
Component-Based SCADA Systems”.
3- B. M. Michelson, ``Event-driven architecture overview,'' Patricia Seybold Group, vol.
2, no. 12, p. 1057, 2006.

Page 4 of 4

You might also like